ABSTRACT Bird diversity is a key indicator of urban ecosystems. While particular focus has traditionally been directed toward the richness and distribution patterns of birds within urban areas, less attention has been paid to the individual spatial–temporal relationships among bird species. Understanding these implicit relationships is crucial for delivering a more comprehensive view on urban biodiversity, relying not only on aggregated statistical indices, but also on the investigation of deeper species‐to‐species similarities in terms of birds' motion behaviors. This study introduces Birds2Vec, a novel approach for analyzing inter‐species behavioral connections based on distributions of bird observations in space and time. Leveraging a data‐driven unsupervised learning model, inspired by the Word2Vec perspective in natural language processing, we intend to construct a multi‐dimensional semantic space whereby each bird species is represented as an embedding vector, with distances between vectors indicating the degree of spatial–temporal relatedness between species. Additionally, by combining bird species' embeddings, a further vector space can be created for urban regions, allowing for investigations on the environmental features that may have a greater impact on the local biodiversity. The overall modeling outcomes have the potential of assessing if and when the spatial distribution of bird species is influenced by the traditional biological taxonomy, namely if phylogenetically‐related birds have common overlapping behaviors in urban environments. Our methodology represents an attempt at adapting advanced concepts and tools of the artificial intelligence sector to the context of biodiversity, offering new perspectives and scientific support for urban planning and conservation efforts.
Crivellari et al. (Mon,) studied this question.